Home | History | Annotate | Download | only in fio
      1 /*
      2  * fio - the flexible io tester
      3  *
      4  * Copyright (C) 2005 Jens Axboe <axboe (at) suse.de>
      5  * Copyright (C) 2006-2012 Jens Axboe <axboe (at) kernel.dk>
      6  *
      7  * The license below covers all files distributed with fio unless otherwise
      8  * noted in the file itself.
      9  *
     10  *  This program is free software; you can redistribute it and/or modify
     11  *  it under the terms of the GNU General Public License version 2 as
     12  *  published by the Free Software Foundation.
     13  *
     14  *  This program is distributed in the hope that it will be useful,
     15  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
     16  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     17  *  GNU General Public License for more details.
     18  *
     19  *  You should have received a copy of the GNU General Public License
     20  *  along with this program; if not, write to the Free Software
     21  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
     22  *
     23  */
     24 #include <unistd.h>
     25 #include <fcntl.h>
     26 #include <string.h>
     27 #include <limits.h>
     28 #include <signal.h>
     29 #include <time.h>
     30 #include <locale.h>
     31 #include <assert.h>
     32 #include <time.h>
     33 #include <inttypes.h>
     34 #include <sys/stat.h>
     35 #include <sys/wait.h>
     36 #include <sys/ipc.h>
     37 #include <sys/mman.h>
     38 #include <math.h>
     39 
     40 #include "fio.h"
     41 #ifndef FIO_NO_HAVE_SHM_H
     42 #include <sys/shm.h>
     43 #endif
     44 #include "hash.h"
     45 #include "smalloc.h"
     46 #include "verify.h"
     47 #include "trim.h"
     48 #include "diskutil.h"
     49 #include "cgroup.h"
     50 #include "profile.h"
     51 #include "lib/rand.h"
     52 #include "lib/memalign.h"
     53 #include "server.h"
     54 #include "lib/getrusage.h"
     55 #include "idletime.h"
     56 #include "err.h"
     57 #include "workqueue.h"
     58 #include "lib/mountcheck.h"
     59 #include "rate-submit.h"
     60 #include "helper_thread.h"
     61 
     62 static struct fio_mutex *startup_mutex;
     63 static struct flist_head *cgroup_list;
     64 static char *cgroup_mnt;
     65 static int exit_value;
     66 static volatile int fio_abort;
     67 static unsigned int nr_process = 0;
     68 static unsigned int nr_thread = 0;
     69 
     70 struct io_log *agg_io_log[DDIR_RWDIR_CNT];
     71 
     72 int groupid = 0;
     73 unsigned int thread_number = 0;
     74 unsigned int stat_number = 0;
     75 int shm_id = 0;
     76 int temp_stall_ts;
     77 unsigned long done_secs = 0;
     78 
     79 #define JOB_START_TIMEOUT	(5 * 1000)
     80 
     81 static void sig_int(int sig)
     82 {
     83 	if (threads) {
     84 		if (is_backend)
     85 			fio_server_got_signal(sig);
     86 		else {
     87 			log_info("\nfio: terminating on signal %d\n", sig);
     88 			log_info_flush();
     89 			exit_value = 128;
     90 		}
     91 
     92 		fio_terminate_threads(TERMINATE_ALL);
     93 	}
     94 }
     95 
     96 void sig_show_status(int sig)
     97 {
     98 	show_running_run_stats();
     99 }
    100 
    101 static void set_sig_handlers(void)
    102 {
    103 	struct sigaction act;
    104 
    105 	memset(&act, 0, sizeof(act));
    106 	act.sa_handler = sig_int;
    107 	act.sa_flags = SA_RESTART;
    108 	sigaction(SIGINT, &act, NULL);
    109 
    110 	memset(&act, 0, sizeof(act));
    111 	act.sa_handler = sig_int;
    112 	act.sa_flags = SA_RESTART;
    113 	sigaction(SIGTERM, &act, NULL);
    114 
    115 /* Windows uses SIGBREAK as a quit signal from other applications */
    116 #ifdef WIN32
    117 	memset(&act, 0, sizeof(act));
    118 	act.sa_handler = sig_int;
    119 	act.sa_flags = SA_RESTART;
    120 	sigaction(SIGBREAK, &act, NULL);
    121 #endif
    122 
    123 	memset(&act, 0, sizeof(act));
    124 	act.sa_handler = sig_show_status;
    125 	act.sa_flags = SA_RESTART;
    126 	sigaction(SIGUSR1, &act, NULL);
    127 
    128 	if (is_backend) {
    129 		memset(&act, 0, sizeof(act));
    130 		act.sa_handler = sig_int;
    131 		act.sa_flags = SA_RESTART;
    132 		sigaction(SIGPIPE, &act, NULL);
    133 	}
    134 }
    135 
    136 /*
    137  * Check if we are above the minimum rate given.
    138  */
    139 static bool __check_min_rate(struct thread_data *td, struct timeval *now,
    140 			     enum fio_ddir ddir)
    141 {
    142 	unsigned long long bytes = 0;
    143 	unsigned long iops = 0;
    144 	unsigned long spent;
    145 	unsigned long rate;
    146 	unsigned int ratemin = 0;
    147 	unsigned int rate_iops = 0;
    148 	unsigned int rate_iops_min = 0;
    149 
    150 	assert(ddir_rw(ddir));
    151 
    152 	if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
    153 		return false;
    154 
    155 	/*
    156 	 * allow a 2 second settle period in the beginning
    157 	 */
    158 	if (mtime_since(&td->start, now) < 2000)
    159 		return false;
    160 
    161 	iops += td->this_io_blocks[ddir];
    162 	bytes += td->this_io_bytes[ddir];
    163 	ratemin += td->o.ratemin[ddir];
    164 	rate_iops += td->o.rate_iops[ddir];
    165 	rate_iops_min += td->o.rate_iops_min[ddir];
    166 
    167 	/*
    168 	 * if rate blocks is set, sample is running
    169 	 */
    170 	if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) {
    171 		spent = mtime_since(&td->lastrate[ddir], now);
    172 		if (spent < td->o.ratecycle)
    173 			return false;
    174 
    175 		if (td->o.rate[ddir] || td->o.ratemin[ddir]) {
    176 			/*
    177 			 * check bandwidth specified rate
    178 			 */
    179 			if (bytes < td->rate_bytes[ddir]) {
    180 				log_err("%s: rate_min=%uB/s not met, only transferred %lluB\n",
    181 					td->o.name, ratemin, bytes);
    182 				return true;
    183 			} else {
    184 				if (spent)
    185 					rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
    186 				else
    187 					rate = 0;
    188 
    189 				if (rate < ratemin ||
    190 				    bytes < td->rate_bytes[ddir]) {
    191 					log_err("%s: rate_min=%uB/s not met, got %luB/s\n",
    192 						td->o.name, ratemin, rate);
    193 					return true;
    194 				}
    195 			}
    196 		} else {
    197 			/*
    198 			 * checks iops specified rate
    199 			 */
    200 			if (iops < rate_iops) {
    201 				log_err("%s: rate_iops_min=%u not met, only performed %lu IOs\n",
    202 						td->o.name, rate_iops, iops);
    203 				return true;
    204 			} else {
    205 				if (spent)
    206 					rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
    207 				else
    208 					rate = 0;
    209 
    210 				if (rate < rate_iops_min ||
    211 				    iops < td->rate_blocks[ddir]) {
    212 					log_err("%s: rate_iops_min=%u not met, got %lu IOPS\n",
    213 						td->o.name, rate_iops_min, rate);
    214 					return true;
    215 				}
    216 			}
    217 		}
    218 	}
    219 
    220 	td->rate_bytes[ddir] = bytes;
    221 	td->rate_blocks[ddir] = iops;
    222 	memcpy(&td->lastrate[ddir], now, sizeof(*now));
    223 	return false;
    224 }
    225 
    226 static bool check_min_rate(struct thread_data *td, struct timeval *now)
    227 {
    228 	bool ret = false;
    229 
    230 	if (td->bytes_done[DDIR_READ])
    231 		ret |= __check_min_rate(td, now, DDIR_READ);
    232 	if (td->bytes_done[DDIR_WRITE])
    233 		ret |= __check_min_rate(td, now, DDIR_WRITE);
    234 	if (td->bytes_done[DDIR_TRIM])
    235 		ret |= __check_min_rate(td, now, DDIR_TRIM);
    236 
    237 	return ret;
    238 }
    239 
    240 /*
    241  * When job exits, we can cancel the in-flight IO if we are using async
    242  * io. Attempt to do so.
    243  */
    244 static void cleanup_pending_aio(struct thread_data *td)
    245 {
    246 	int r;
    247 
    248 	/*
    249 	 * get immediately available events, if any
    250 	 */
    251 	r = io_u_queued_complete(td, 0);
    252 	if (r < 0)
    253 		return;
    254 
    255 	/*
    256 	 * now cancel remaining active events
    257 	 */
    258 	if (td->io_ops->cancel) {
    259 		struct io_u *io_u;
    260 		int i;
    261 
    262 		io_u_qiter(&td->io_u_all, io_u, i) {
    263 			if (io_u->flags & IO_U_F_FLIGHT) {
    264 				r = td->io_ops->cancel(td, io_u);
    265 				if (!r)
    266 					put_io_u(td, io_u);
    267 			}
    268 		}
    269 	}
    270 
    271 	if (td->cur_depth)
    272 		r = io_u_queued_complete(td, td->cur_depth);
    273 }
    274 
    275 /*
    276  * Helper to handle the final sync of a file. Works just like the normal
    277  * io path, just does everything sync.
    278  */
    279 static bool fio_io_sync(struct thread_data *td, struct fio_file *f)
    280 {
    281 	struct io_u *io_u = __get_io_u(td);
    282 	int ret;
    283 
    284 	if (!io_u)
    285 		return true;
    286 
    287 	io_u->ddir = DDIR_SYNC;
    288 	io_u->file = f;
    289 
    290 	if (td_io_prep(td, io_u)) {
    291 		put_io_u(td, io_u);
    292 		return true;
    293 	}
    294 
    295 requeue:
    296 	ret = td_io_queue(td, io_u);
    297 	if (ret < 0) {
    298 		td_verror(td, io_u->error, "td_io_queue");
    299 		put_io_u(td, io_u);
    300 		return true;
    301 	} else if (ret == FIO_Q_QUEUED) {
    302 		if (td_io_commit(td))
    303 			return true;
    304 		if (io_u_queued_complete(td, 1) < 0)
    305 			return true;
    306 	} else if (ret == FIO_Q_COMPLETED) {
    307 		if (io_u->error) {
    308 			td_verror(td, io_u->error, "td_io_queue");
    309 			return true;
    310 		}
    311 
    312 		if (io_u_sync_complete(td, io_u) < 0)
    313 			return true;
    314 	} else if (ret == FIO_Q_BUSY) {
    315 		if (td_io_commit(td))
    316 			return true;
    317 		goto requeue;
    318 	}
    319 
    320 	return false;
    321 }
    322 
    323 static int fio_file_fsync(struct thread_data *td, struct fio_file *f)
    324 {
    325 	int ret;
    326 
    327 	if (fio_file_open(f))
    328 		return fio_io_sync(td, f);
    329 
    330 	if (td_io_open_file(td, f))
    331 		return 1;
    332 
    333 	ret = fio_io_sync(td, f);
    334 	td_io_close_file(td, f);
    335 	return ret;
    336 }
    337 
    338 static inline void __update_tv_cache(struct thread_data *td)
    339 {
    340 	fio_gettime(&td->tv_cache, NULL);
    341 }
    342 
    343 static inline void update_tv_cache(struct thread_data *td)
    344 {
    345 	if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask)
    346 		__update_tv_cache(td);
    347 }
    348 
    349 static inline bool runtime_exceeded(struct thread_data *td, struct timeval *t)
    350 {
    351 	if (in_ramp_time(td))
    352 		return false;
    353 	if (!td->o.timeout)
    354 		return false;
    355 	if (utime_since(&td->epoch, t) >= td->o.timeout)
    356 		return true;
    357 
    358 	return false;
    359 }
    360 
    361 /*
    362  * We need to update the runtime consistently in ms, but keep a running
    363  * tally of the current elapsed time in microseconds for sub millisecond
    364  * updates.
    365  */
    366 static inline void update_runtime(struct thread_data *td,
    367 				  unsigned long long *elapsed_us,
    368 				  const enum fio_ddir ddir)
    369 {
    370 	if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only)
    371 		return;
    372 
    373 	td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000;
    374 	elapsed_us[ddir] += utime_since_now(&td->start);
    375 	td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000;
    376 }
    377 
    378 static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
    379 				int *retptr)
    380 {
    381 	int ret = *retptr;
    382 
    383 	if (ret < 0 || td->error) {
    384 		int err = td->error;
    385 		enum error_type_bit eb;
    386 
    387 		if (ret < 0)
    388 			err = -ret;
    389 
    390 		eb = td_error_type(ddir, err);
    391 		if (!(td->o.continue_on_error & (1 << eb)))
    392 			return true;
    393 
    394 		if (td_non_fatal_error(td, eb, err)) {
    395 		        /*
    396 		         * Continue with the I/Os in case of
    397 			 * a non fatal error.
    398 			 */
    399 			update_error_count(td, err);
    400 			td_clear_error(td);
    401 			*retptr = 0;
    402 			return false;
    403 		} else if (td->o.fill_device && err == ENOSPC) {
    404 			/*
    405 			 * We expect to hit this error if
    406 			 * fill_device option is set.
    407 			 */
    408 			td_clear_error(td);
    409 			fio_mark_td_terminate(td);
    410 			return true;
    411 		} else {
    412 			/*
    413 			 * Stop the I/O in case of a fatal
    414 			 * error.
    415 			 */
    416 			update_error_count(td, err);
    417 			return true;
    418 		}
    419 	}
    420 
    421 	return false;
    422 }
    423 
    424 static void check_update_rusage(struct thread_data *td)
    425 {
    426 	if (td->update_rusage) {
    427 		td->update_rusage = 0;
    428 		update_rusage_stat(td);
    429 		fio_mutex_up(td->rusage_sem);
    430 	}
    431 }
    432 
    433 static int wait_for_completions(struct thread_data *td, struct timeval *time)
    434 {
    435 	const int full = queue_full(td);
    436 	int min_evts = 0;
    437 	int ret;
    438 
    439 	if (td->flags & TD_F_REGROW_LOGS)
    440 		return io_u_quiesce(td);
    441 
    442 	/*
    443 	 * if the queue is full, we MUST reap at least 1 event
    444 	 */
    445 	min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth);
    446 	if ((full && !min_evts) || !td->o.iodepth_batch_complete_min)
    447 		min_evts = 1;
    448 
    449 	if (time && (__should_check_rate(td, DDIR_READ) ||
    450 	    __should_check_rate(td, DDIR_WRITE) ||
    451 	    __should_check_rate(td, DDIR_TRIM)))
    452 		fio_gettime(time, NULL);
    453 
    454 	do {
    455 		ret = io_u_queued_complete(td, min_evts);
    456 		if (ret < 0)
    457 			break;
    458 	} while (full && (td->cur_depth > td->o.iodepth_low));
    459 
    460 	return ret;
    461 }
    462 
    463 int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
    464 		   enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
    465 		   struct timeval *comp_time)
    466 {
    467 	int ret2;
    468 
    469 	switch (*ret) {
    470 	case FIO_Q_COMPLETED:
    471 		if (io_u->error) {
    472 			*ret = -io_u->error;
    473 			clear_io_u(td, io_u);
    474 		} else if (io_u->resid) {
    475 			int bytes = io_u->xfer_buflen - io_u->resid;
    476 			struct fio_file *f = io_u->file;
    477 
    478 			if (bytes_issued)
    479 				*bytes_issued += bytes;
    480 
    481 			if (!from_verify)
    482 				trim_io_piece(td, io_u);
    483 
    484 			/*
    485 			 * zero read, fail
    486 			 */
    487 			if (!bytes) {
    488 				if (!from_verify)
    489 					unlog_io_piece(td, io_u);
    490 				td_verror(td, EIO, "full resid");
    491 				put_io_u(td, io_u);
    492 				break;
    493 			}
    494 
    495 			io_u->xfer_buflen = io_u->resid;
    496 			io_u->xfer_buf += bytes;
    497 			io_u->offset += bytes;
    498 
    499 			if (ddir_rw(io_u->ddir))
    500 				td->ts.short_io_u[io_u->ddir]++;
    501 
    502 			f = io_u->file;
    503 			if (io_u->offset == f->real_file_size)
    504 				goto sync_done;
    505 
    506 			requeue_io_u(td, &io_u);
    507 		} else {
    508 sync_done:
    509 			if (comp_time && (__should_check_rate(td, DDIR_READ) ||
    510 			    __should_check_rate(td, DDIR_WRITE) ||
    511 			    __should_check_rate(td, DDIR_TRIM)))
    512 				fio_gettime(comp_time, NULL);
    513 
    514 			*ret = io_u_sync_complete(td, io_u);
    515 			if (*ret < 0)
    516 				break;
    517 		}
    518 
    519 		if (td->flags & TD_F_REGROW_LOGS)
    520 			regrow_logs(td);
    521 
    522 		/*
    523 		 * when doing I/O (not when verifying),
    524 		 * check for any errors that are to be ignored
    525 		 */
    526 		if (!from_verify)
    527 			break;
    528 
    529 		return 0;
    530 	case FIO_Q_QUEUED:
    531 		/*
    532 		 * if the engine doesn't have a commit hook,
    533 		 * the io_u is really queued. if it does have such
    534 		 * a hook, it has to call io_u_queued() itself.
    535 		 */
    536 		if (td->io_ops->commit == NULL)
    537 			io_u_queued(td, io_u);
    538 		if (bytes_issued)
    539 			*bytes_issued += io_u->xfer_buflen;
    540 		break;
    541 	case FIO_Q_BUSY:
    542 		if (!from_verify)
    543 			unlog_io_piece(td, io_u);
    544 		requeue_io_u(td, &io_u);
    545 		ret2 = td_io_commit(td);
    546 		if (ret2 < 0)
    547 			*ret = ret2;
    548 		break;
    549 	default:
    550 		assert(*ret < 0);
    551 		td_verror(td, -(*ret), "td_io_queue");
    552 		break;
    553 	}
    554 
    555 	if (break_on_this_error(td, ddir, ret))
    556 		return 1;
    557 
    558 	return 0;
    559 }
    560 
    561 static inline bool io_in_polling(struct thread_data *td)
    562 {
    563 	return !td->o.iodepth_batch_complete_min &&
    564 		   !td->o.iodepth_batch_complete_max;
    565 }
    566 /*
    567  * Unlinks files from thread data fio_file structure
    568  */
    569 static int unlink_all_files(struct thread_data *td)
    570 {
    571 	struct fio_file *f;
    572 	unsigned int i;
    573 	int ret = 0;
    574 
    575 	for_each_file(td, f, i) {
    576 		if (f->filetype != FIO_TYPE_FILE)
    577 			continue;
    578 		ret = td_io_unlink_file(td, f);
    579 		if (ret)
    580 			break;
    581 	}
    582 
    583 	if (ret)
    584 		td_verror(td, ret, "unlink_all_files");
    585 
    586 	return ret;
    587 }
    588 
    589 /*
    590  * The main verify engine. Runs over the writes we previously submitted,
    591  * reads the blocks back in, and checks the crc/md5 of the data.
    592  */
    593 static void do_verify(struct thread_data *td, uint64_t verify_bytes)
    594 {
    595 	struct fio_file *f;
    596 	struct io_u *io_u;
    597 	int ret, min_events;
    598 	unsigned int i;
    599 
    600 	dprint(FD_VERIFY, "starting loop\n");
    601 
    602 	/*
    603 	 * sync io first and invalidate cache, to make sure we really
    604 	 * read from disk.
    605 	 */
    606 	for_each_file(td, f, i) {
    607 		if (!fio_file_open(f))
    608 			continue;
    609 		if (fio_io_sync(td, f))
    610 			break;
    611 		if (file_invalidate_cache(td, f))
    612 			break;
    613 	}
    614 
    615 	check_update_rusage(td);
    616 
    617 	if (td->error)
    618 		return;
    619 
    620 	/*
    621 	 * verify_state needs to be reset before verification
    622 	 * proceeds so that expected random seeds match actual
    623 	 * random seeds in headers. The main loop will reset
    624 	 * all random number generators if randrepeat is set.
    625 	 */
    626 	if (!td->o.rand_repeatable)
    627 		td_fill_verify_state_seed(td);
    628 
    629 	td_set_runstate(td, TD_VERIFYING);
    630 
    631 	io_u = NULL;
    632 	while (!td->terminate) {
    633 		enum fio_ddir ddir;
    634 		int full;
    635 
    636 		update_tv_cache(td);
    637 		check_update_rusage(td);
    638 
    639 		if (runtime_exceeded(td, &td->tv_cache)) {
    640 			__update_tv_cache(td);
    641 			if (runtime_exceeded(td, &td->tv_cache)) {
    642 				fio_mark_td_terminate(td);
    643 				break;
    644 			}
    645 		}
    646 
    647 		if (flow_threshold_exceeded(td))
    648 			continue;
    649 
    650 		if (!td->o.experimental_verify) {
    651 			io_u = __get_io_u(td);
    652 			if (!io_u)
    653 				break;
    654 
    655 			if (get_next_verify(td, io_u)) {
    656 				put_io_u(td, io_u);
    657 				break;
    658 			}
    659 
    660 			if (td_io_prep(td, io_u)) {
    661 				put_io_u(td, io_u);
    662 				break;
    663 			}
    664 		} else {
    665 			if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes)
    666 				break;
    667 
    668 			while ((io_u = get_io_u(td)) != NULL) {
    669 				if (IS_ERR_OR_NULL(io_u)) {
    670 					io_u = NULL;
    671 					ret = FIO_Q_BUSY;
    672 					goto reap;
    673 				}
    674 
    675 				/*
    676 				 * We are only interested in the places where
    677 				 * we wrote or trimmed IOs. Turn those into
    678 				 * reads for verification purposes.
    679 				 */
    680 				if (io_u->ddir == DDIR_READ) {
    681 					/*
    682 					 * Pretend we issued it for rwmix
    683 					 * accounting
    684 					 */
    685 					td->io_issues[DDIR_READ]++;
    686 					put_io_u(td, io_u);
    687 					continue;
    688 				} else if (io_u->ddir == DDIR_TRIM) {
    689 					io_u->ddir = DDIR_READ;
    690 					io_u_set(td, io_u, IO_U_F_TRIMMED);
    691 					break;
    692 				} else if (io_u->ddir == DDIR_WRITE) {
    693 					io_u->ddir = DDIR_READ;
    694 					break;
    695 				} else {
    696 					put_io_u(td, io_u);
    697 					continue;
    698 				}
    699 			}
    700 
    701 			if (!io_u)
    702 				break;
    703 		}
    704 
    705 		if (verify_state_should_stop(td, io_u)) {
    706 			put_io_u(td, io_u);
    707 			break;
    708 		}
    709 
    710 		if (td->o.verify_async)
    711 			io_u->end_io = verify_io_u_async;
    712 		else
    713 			io_u->end_io = verify_io_u;
    714 
    715 		ddir = io_u->ddir;
    716 		if (!td->o.disable_slat)
    717 			fio_gettime(&io_u->start_time, NULL);
    718 
    719 		ret = td_io_queue(td, io_u);
    720 
    721 		if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL))
    722 			break;
    723 
    724 		/*
    725 		 * if we can queue more, do so. but check if there are
    726 		 * completed io_u's first. Note that we can get BUSY even
    727 		 * without IO queued, if the system is resource starved.
    728 		 */
    729 reap:
    730 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
    731 		if (full || io_in_polling(td))
    732 			ret = wait_for_completions(td, NULL);
    733 
    734 		if (ret < 0)
    735 			break;
    736 	}
    737 
    738 	check_update_rusage(td);
    739 
    740 	if (!td->error) {
    741 		min_events = td->cur_depth;
    742 
    743 		if (min_events)
    744 			ret = io_u_queued_complete(td, min_events);
    745 	} else
    746 		cleanup_pending_aio(td);
    747 
    748 	td_set_runstate(td, TD_RUNNING);
    749 
    750 	dprint(FD_VERIFY, "exiting loop\n");
    751 }
    752 
    753 static bool exceeds_number_ios(struct thread_data *td)
    754 {
    755 	unsigned long long number_ios;
    756 
    757 	if (!td->o.number_ios)
    758 		return false;
    759 
    760 	number_ios = ddir_rw_sum(td->io_blocks);
    761 	number_ios += td->io_u_queued + td->io_u_in_flight;
    762 
    763 	return number_ios >= (td->o.number_ios * td->loops);
    764 }
    765 
    766 static bool io_bytes_exceeded(struct thread_data *td, uint64_t *this_bytes)
    767 {
    768 	unsigned long long bytes, limit;
    769 
    770 	if (td_rw(td))
    771 		bytes = this_bytes[DDIR_READ] + this_bytes[DDIR_WRITE];
    772 	else if (td_write(td))
    773 		bytes = this_bytes[DDIR_WRITE];
    774 	else if (td_read(td))
    775 		bytes = this_bytes[DDIR_READ];
    776 	else
    777 		bytes = this_bytes[DDIR_TRIM];
    778 
    779 	if (td->o.io_size)
    780 		limit = td->o.io_size;
    781 	else
    782 		limit = td->o.size;
    783 
    784 	limit *= td->loops;
    785 	return bytes >= limit || exceeds_number_ios(td);
    786 }
    787 
    788 static bool io_issue_bytes_exceeded(struct thread_data *td)
    789 {
    790 	return io_bytes_exceeded(td, td->io_issue_bytes);
    791 }
    792 
    793 static bool io_complete_bytes_exceeded(struct thread_data *td)
    794 {
    795 	return io_bytes_exceeded(td, td->this_io_bytes);
    796 }
    797 
    798 /*
    799  * used to calculate the next io time for rate control
    800  *
    801  */
    802 static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
    803 {
    804 	uint64_t secs, remainder, bps, bytes, iops;
    805 
    806 	assert(!(td->flags & TD_F_CHILD));
    807 	bytes = td->rate_io_issue_bytes[ddir];
    808 	bps = td->rate_bps[ddir];
    809 
    810 	if (td->o.rate_process == RATE_PROCESS_POISSON) {
    811 		uint64_t val;
    812 		iops = bps / td->o.bs[ddir];
    813 		val = (int64_t) (1000000 / iops) *
    814 				-logf(__rand_0_1(&td->poisson_state[ddir]));
    815 		if (val) {
    816 			dprint(FD_RATE, "poisson rate iops=%llu, ddir=%d\n",
    817 					(unsigned long long) 1000000 / val,
    818 					ddir);
    819 		}
    820 		td->last_usec[ddir] += val;
    821 		return td->last_usec[ddir];
    822 	} else if (bps) {
    823 		secs = bytes / bps;
    824 		remainder = bytes % bps;
    825 		return remainder * 1000000 / bps + secs * 1000000;
    826 	}
    827 
    828 	return 0;
    829 }
    830 
    831 /*
    832  * Main IO worker function. It retrieves io_u's to process and queues
    833  * and reaps them, checking for rate and errors along the way.
    834  *
    835  * Returns number of bytes written and trimmed.
    836  */
    837 static void do_io(struct thread_data *td, uint64_t *bytes_done)
    838 {
    839 	unsigned int i;
    840 	int ret = 0;
    841 	uint64_t total_bytes, bytes_issued = 0;
    842 
    843 	for (i = 0; i < DDIR_RWDIR_CNT; i++)
    844 		bytes_done[i] = td->bytes_done[i];
    845 
    846 	if (in_ramp_time(td))
    847 		td_set_runstate(td, TD_RAMP);
    848 	else
    849 		td_set_runstate(td, TD_RUNNING);
    850 
    851 	lat_target_init(td);
    852 
    853 	total_bytes = td->o.size;
    854 	/*
    855 	* Allow random overwrite workloads to write up to io_size
    856 	* before starting verification phase as 'size' doesn't apply.
    857 	*/
    858 	if (td_write(td) && td_random(td) && td->o.norandommap)
    859 		total_bytes = max(total_bytes, (uint64_t) td->o.io_size);
    860 	/*
    861 	 * If verify_backlog is enabled, we'll run the verify in this
    862 	 * handler as well. For that case, we may need up to twice the
    863 	 * amount of bytes.
    864 	 */
    865 	if (td->o.verify != VERIFY_NONE &&
    866 	   (td_write(td) && td->o.verify_backlog))
    867 		total_bytes += td->o.size;
    868 
    869 	/* In trimwrite mode, each byte is trimmed and then written, so
    870 	 * allow total_bytes to be twice as big */
    871 	if (td_trimwrite(td))
    872 		total_bytes += td->total_io_size;
    873 
    874 	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
    875 		(!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) ||
    876 		td->o.time_based) {
    877 		struct timeval comp_time;
    878 		struct io_u *io_u;
    879 		int full;
    880 		enum fio_ddir ddir;
    881 
    882 		check_update_rusage(td);
    883 
    884 		if (td->terminate || td->done)
    885 			break;
    886 
    887 		update_tv_cache(td);
    888 
    889 		if (runtime_exceeded(td, &td->tv_cache)) {
    890 			__update_tv_cache(td);
    891 			if (runtime_exceeded(td, &td->tv_cache)) {
    892 				fio_mark_td_terminate(td);
    893 				break;
    894 			}
    895 		}
    896 
    897 		if (flow_threshold_exceeded(td))
    898 			continue;
    899 
    900 		/*
    901 		 * Break if we exceeded the bytes. The exception is time
    902 		 * based runs, but we still need to break out of the loop
    903 		 * for those to run verification, if enabled.
    904 		 */
    905 		if (bytes_issued >= total_bytes &&
    906 		    (!td->o.time_based ||
    907 		     (td->o.time_based && td->o.verify != VERIFY_NONE)))
    908 			break;
    909 
    910 		io_u = get_io_u(td);
    911 		if (IS_ERR_OR_NULL(io_u)) {
    912 			int err = PTR_ERR(io_u);
    913 
    914 			io_u = NULL;
    915 			if (err == -EBUSY) {
    916 				ret = FIO_Q_BUSY;
    917 				goto reap;
    918 			}
    919 			if (td->o.latency_target)
    920 				goto reap;
    921 			break;
    922 		}
    923 
    924 		ddir = io_u->ddir;
    925 
    926 		/*
    927 		 * Add verification end_io handler if:
    928 		 *	- Asked to verify (!td_rw(td))
    929 		 *	- Or the io_u is from our verify list (mixed write/ver)
    930 		 */
    931 		if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
    932 		    ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
    933 
    934 			if (!td->o.verify_pattern_bytes) {
    935 				io_u->rand_seed = __rand(&td->verify_state);
    936 				if (sizeof(int) != sizeof(long *))
    937 					io_u->rand_seed *= __rand(&td->verify_state);
    938 			}
    939 
    940 			if (verify_state_should_stop(td, io_u)) {
    941 				put_io_u(td, io_u);
    942 				break;
    943 			}
    944 
    945 			if (td->o.verify_async)
    946 				io_u->end_io = verify_io_u_async;
    947 			else
    948 				io_u->end_io = verify_io_u;
    949 			td_set_runstate(td, TD_VERIFYING);
    950 		} else if (in_ramp_time(td))
    951 			td_set_runstate(td, TD_RAMP);
    952 		else
    953 			td_set_runstate(td, TD_RUNNING);
    954 
    955 		/*
    956 		 * Always log IO before it's issued, so we know the specific
    957 		 * order of it. The logged unit will track when the IO has
    958 		 * completed.
    959 		 */
    960 		if (td_write(td) && io_u->ddir == DDIR_WRITE &&
    961 		    td->o.do_verify &&
    962 		    td->o.verify != VERIFY_NONE &&
    963 		    !td->o.experimental_verify)
    964 			log_io_piece(td, io_u);
    965 
    966 		if (td->o.io_submit_mode == IO_MODE_OFFLOAD) {
    967 			const unsigned long blen = io_u->xfer_buflen;
    968 			const enum fio_ddir ddir = acct_ddir(io_u);
    969 
    970 			if (td->error)
    971 				break;
    972 
    973 			workqueue_enqueue(&td->io_wq, &io_u->work);
    974 			ret = FIO_Q_QUEUED;
    975 
    976 			if (ddir_rw(ddir)) {
    977 				td->io_issues[ddir]++;
    978 				td->io_issue_bytes[ddir] += blen;
    979 				td->rate_io_issue_bytes[ddir] += blen;
    980 			}
    981 
    982 			if (should_check_rate(td))
    983 				td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
    984 
    985 		} else {
    986 			ret = td_io_queue(td, io_u);
    987 
    988 			if (should_check_rate(td))
    989 				td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
    990 
    991 			if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time))
    992 				break;
    993 
    994 			/*
    995 			 * See if we need to complete some commands. Note that
    996 			 * we can get BUSY even without IO queued, if the
    997 			 * system is resource starved.
    998 			 */
    999 reap:
   1000 			full = queue_full(td) ||
   1001 				(ret == FIO_Q_BUSY && td->cur_depth);
   1002 			if (full || io_in_polling(td))
   1003 				ret = wait_for_completions(td, &comp_time);
   1004 		}
   1005 		if (ret < 0)
   1006 			break;
   1007 		if (!ddir_rw_sum(td->bytes_done) &&
   1008 		    !td_ioengine_flagged(td, FIO_NOIO))
   1009 			continue;
   1010 
   1011 		if (!in_ramp_time(td) && should_check_rate(td)) {
   1012 			if (check_min_rate(td, &comp_time)) {
   1013 				if (exitall_on_terminate || td->o.exitall_error)
   1014 					fio_terminate_threads(td->groupid);
   1015 				td_verror(td, EIO, "check_min_rate");
   1016 				break;
   1017 			}
   1018 		}
   1019 		if (!in_ramp_time(td) && td->o.latency_target)
   1020 			lat_target_check(td);
   1021 
   1022 		if (td->o.thinktime) {
   1023 			unsigned long long b;
   1024 
   1025 			b = ddir_rw_sum(td->io_blocks);
   1026 			if (!(b % td->o.thinktime_blocks)) {
   1027 				int left;
   1028 
   1029 				io_u_quiesce(td);
   1030 
   1031 				if (td->o.thinktime_spin)
   1032 					usec_spin(td->o.thinktime_spin);
   1033 
   1034 				left = td->o.thinktime - td->o.thinktime_spin;
   1035 				if (left)
   1036 					usec_sleep(td, left);
   1037 			}
   1038 		}
   1039 	}
   1040 
   1041 	check_update_rusage(td);
   1042 
   1043 	if (td->trim_entries)
   1044 		log_err("fio: %lu trim entries leaked?\n", td->trim_entries);
   1045 
   1046 	if (td->o.fill_device && td->error == ENOSPC) {
   1047 		td->error = 0;
   1048 		fio_mark_td_terminate(td);
   1049 	}
   1050 	if (!td->error) {
   1051 		struct fio_file *f;
   1052 
   1053 		if (td->o.io_submit_mode == IO_MODE_OFFLOAD) {
   1054 			workqueue_flush(&td->io_wq);
   1055 			i = 0;
   1056 		} else
   1057 			i = td->cur_depth;
   1058 
   1059 		if (i) {
   1060 			ret = io_u_queued_complete(td, i);
   1061 			if (td->o.fill_device && td->error == ENOSPC)
   1062 				td->error = 0;
   1063 		}
   1064 
   1065 		if (should_fsync(td) && td->o.end_fsync) {
   1066 			td_set_runstate(td, TD_FSYNCING);
   1067 
   1068 			for_each_file(td, f, i) {
   1069 				if (!fio_file_fsync(td, f))
   1070 					continue;
   1071 
   1072 				log_err("fio: end_fsync failed for file %s\n",
   1073 								f->file_name);
   1074 			}
   1075 		}
   1076 	} else
   1077 		cleanup_pending_aio(td);
   1078 
   1079 	/*
   1080 	 * stop job if we failed doing any IO
   1081 	 */
   1082 	if (!ddir_rw_sum(td->this_io_bytes))
   1083 		td->done = 1;
   1084 
   1085 	for (i = 0; i < DDIR_RWDIR_CNT; i++)
   1086 		bytes_done[i] = td->bytes_done[i] - bytes_done[i];
   1087 }
   1088 
   1089 static void free_file_completion_logging(struct thread_data *td)
   1090 {
   1091 	struct fio_file *f;
   1092 	unsigned int i;
   1093 
   1094 	for_each_file(td, f, i) {
   1095 		if (!f->last_write_comp)
   1096 			break;
   1097 		sfree(f->last_write_comp);
   1098 	}
   1099 }
   1100 
   1101 static int init_file_completion_logging(struct thread_data *td,
   1102 					unsigned int depth)
   1103 {
   1104 	struct fio_file *f;
   1105 	unsigned int i;
   1106 
   1107 	if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save)
   1108 		return 0;
   1109 
   1110 	for_each_file(td, f, i) {
   1111 		f->last_write_comp = scalloc(depth, sizeof(uint64_t));
   1112 		if (!f->last_write_comp)
   1113 			goto cleanup;
   1114 	}
   1115 
   1116 	return 0;
   1117 
   1118 cleanup:
   1119 	free_file_completion_logging(td);
   1120 	log_err("fio: failed to alloc write comp data\n");
   1121 	return 1;
   1122 }
   1123 
   1124 static void cleanup_io_u(struct thread_data *td)
   1125 {
   1126 	struct io_u *io_u;
   1127 
   1128 	while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) {
   1129 
   1130 		if (td->io_ops->io_u_free)
   1131 			td->io_ops->io_u_free(td, io_u);
   1132 
   1133 		fio_memfree(io_u, sizeof(*io_u));
   1134 	}
   1135 
   1136 	free_io_mem(td);
   1137 
   1138 	io_u_rexit(&td->io_u_requeues);
   1139 	io_u_qexit(&td->io_u_freelist);
   1140 	io_u_qexit(&td->io_u_all);
   1141 
   1142 	free_file_completion_logging(td);
   1143 }
   1144 
   1145 static int init_io_u(struct thread_data *td)
   1146 {
   1147 	struct io_u *io_u;
   1148 	unsigned int max_bs, min_write;
   1149 	int cl_align, i, max_units;
   1150 	int data_xfer = 1, err;
   1151 	char *p;
   1152 
   1153 	max_units = td->o.iodepth;
   1154 	max_bs = td_max_bs(td);
   1155 	min_write = td->o.min_bs[DDIR_WRITE];
   1156 	td->orig_buffer_size = (unsigned long long) max_bs
   1157 					* (unsigned long long) max_units;
   1158 
   1159 	if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td)))
   1160 		data_xfer = 0;
   1161 
   1162 	err = 0;
   1163 	err += io_u_rinit(&td->io_u_requeues, td->o.iodepth);
   1164 	err += io_u_qinit(&td->io_u_freelist, td->o.iodepth);
   1165 	err += io_u_qinit(&td->io_u_all, td->o.iodepth);
   1166 
   1167 	if (err) {
   1168 		log_err("fio: failed setting up IO queues\n");
   1169 		return 1;
   1170 	}
   1171 
   1172 	/*
   1173 	 * if we may later need to do address alignment, then add any
   1174 	 * possible adjustment here so that we don't cause a buffer
   1175 	 * overflow later. this adjustment may be too much if we get
   1176 	 * lucky and the allocator gives us an aligned address.
   1177 	 */
   1178 	if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
   1179 	    td_ioengine_flagged(td, FIO_RAWIO))
   1180 		td->orig_buffer_size += page_mask + td->o.mem_align;
   1181 
   1182 	if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
   1183 		unsigned long bs;
   1184 
   1185 		bs = td->orig_buffer_size + td->o.hugepage_size - 1;
   1186 		td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1);
   1187 	}
   1188 
   1189 	if (td->orig_buffer_size != (size_t) td->orig_buffer_size) {
   1190 		log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
   1191 		return 1;
   1192 	}
   1193 
   1194 	if (data_xfer && allocate_io_mem(td))
   1195 		return 1;
   1196 
   1197 	if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
   1198 	    td_ioengine_flagged(td, FIO_RAWIO))
   1199 		p = PTR_ALIGN(td->orig_buffer, page_mask) + td->o.mem_align;
   1200 	else
   1201 		p = td->orig_buffer;
   1202 
   1203 	cl_align = os_cache_line_size();
   1204 
   1205 	for (i = 0; i < max_units; i++) {
   1206 		void *ptr;
   1207 
   1208 		if (td->terminate)
   1209 			return 1;
   1210 
   1211 		ptr = fio_memalign(cl_align, sizeof(*io_u));
   1212 		if (!ptr) {
   1213 			log_err("fio: unable to allocate aligned memory\n");
   1214 			break;
   1215 		}
   1216 
   1217 		io_u = ptr;
   1218 		memset(io_u, 0, sizeof(*io_u));
   1219 		INIT_FLIST_HEAD(&io_u->verify_list);
   1220 		dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i);
   1221 
   1222 		if (data_xfer) {
   1223 			io_u->buf = p;
   1224 			dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf);
   1225 
   1226 			if (td_write(td))
   1227 				io_u_fill_buffer(td, io_u, min_write, max_bs);
   1228 			if (td_write(td) && td->o.verify_pattern_bytes) {
   1229 				/*
   1230 				 * Fill the buffer with the pattern if we are
   1231 				 * going to be doing writes.
   1232 				 */
   1233 				fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0);
   1234 			}
   1235 		}
   1236 
   1237 		io_u->index = i;
   1238 		io_u->flags = IO_U_F_FREE;
   1239 		io_u_qpush(&td->io_u_freelist, io_u);
   1240 
   1241 		/*
   1242 		 * io_u never leaves this stack, used for iteration of all
   1243 		 * io_u buffers.
   1244 		 */
   1245 		io_u_qpush(&td->io_u_all, io_u);
   1246 
   1247 		if (td->io_ops->io_u_init) {
   1248 			int ret = td->io_ops->io_u_init(td, io_u);
   1249 
   1250 			if (ret) {
   1251 				log_err("fio: failed to init engine data: %d\n", ret);
   1252 				return 1;
   1253 			}
   1254 		}
   1255 
   1256 		p += max_bs;
   1257 	}
   1258 
   1259 	if (init_file_completion_logging(td, max_units))
   1260 		return 1;
   1261 
   1262 	return 0;
   1263 }
   1264 
   1265 /*
   1266  * This function is Linux specific.
   1267  * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux.
   1268  */
   1269 static int switch_ioscheduler(struct thread_data *td)
   1270 {
   1271 #ifdef FIO_HAVE_IOSCHED_SWITCH
   1272 	char tmp[256], tmp2[128];
   1273 	FILE *f;
   1274 	int ret;
   1275 
   1276 	if (td_ioengine_flagged(td, FIO_DISKLESSIO))
   1277 		return 0;
   1278 
   1279 	assert(td->files && td->files[0]);
   1280 	sprintf(tmp, "%s/queue/scheduler", td->files[0]->du->sysfs_root);
   1281 
   1282 	f = fopen(tmp, "r+");
   1283 	if (!f) {
   1284 		if (errno == ENOENT) {
   1285 			log_err("fio: os or kernel doesn't support IO scheduler"
   1286 				" switching\n");
   1287 			return 0;
   1288 		}
   1289 		td_verror(td, errno, "fopen iosched");
   1290 		return 1;
   1291 	}
   1292 
   1293 	/*
   1294 	 * Set io scheduler.
   1295 	 */
   1296 	ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f);
   1297 	if (ferror(f) || ret != 1) {
   1298 		td_verror(td, errno, "fwrite");
   1299 		fclose(f);
   1300 		return 1;
   1301 	}
   1302 
   1303 	rewind(f);
   1304 
   1305 	/*
   1306 	 * Read back and check that the selected scheduler is now the default.
   1307 	 */
   1308 	memset(tmp, 0, sizeof(tmp));
   1309 	ret = fread(tmp, sizeof(tmp), 1, f);
   1310 	if (ferror(f) || ret < 0) {
   1311 		td_verror(td, errno, "fread");
   1312 		fclose(f);
   1313 		return 1;
   1314 	}
   1315 	/*
   1316 	 * either a list of io schedulers or "none\n" is expected.
   1317 	 */
   1318 	tmp[strlen(tmp) - 1] = '\0';
   1319 
   1320 	/*
   1321 	 * Write to "none" entry doesn't fail, so check the result here.
   1322 	 */
   1323 	if (!strcmp(tmp, "none")) {
   1324 		log_err("fio: io scheduler is not tunable\n");
   1325 		fclose(f);
   1326 		return 0;
   1327 	}
   1328 
   1329 	sprintf(tmp2, "[%s]", td->o.ioscheduler);
   1330 	if (!strstr(tmp, tmp2)) {
   1331 		log_err("fio: io scheduler %s not found\n", td->o.ioscheduler);
   1332 		td_verror(td, EINVAL, "iosched_switch");
   1333 		fclose(f);
   1334 		return 1;
   1335 	}
   1336 
   1337 	fclose(f);
   1338 	return 0;
   1339 #else
   1340 	return 0;
   1341 #endif
   1342 }
   1343 
   1344 static bool keep_running(struct thread_data *td)
   1345 {
   1346 	unsigned long long limit;
   1347 
   1348 	if (td->done)
   1349 		return false;
   1350 	if (td->o.time_based)
   1351 		return true;
   1352 	if (td->o.loops) {
   1353 		td->o.loops--;
   1354 		return true;
   1355 	}
   1356 	if (exceeds_number_ios(td))
   1357 		return false;
   1358 
   1359 	if (td->o.io_size)
   1360 		limit = td->o.io_size;
   1361 	else
   1362 		limit = td->o.size;
   1363 
   1364 	if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) {
   1365 		uint64_t diff;
   1366 
   1367 		/*
   1368 		 * If the difference is less than the maximum IO size, we
   1369 		 * are done.
   1370 		 */
   1371 		diff = limit - ddir_rw_sum(td->io_bytes);
   1372 		if (diff < td_max_bs(td))
   1373 			return false;
   1374 
   1375 		if (fio_files_done(td) && !td->o.io_size)
   1376 			return false;
   1377 
   1378 		return true;
   1379 	}
   1380 
   1381 	return false;
   1382 }
   1383 
   1384 static int exec_string(struct thread_options *o, const char *string, const char *mode)
   1385 {
   1386 	size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
   1387 	int ret;
   1388 	char *str;
   1389 
   1390 	str = malloc(newlen);
   1391 	sprintf(str, "%s &> %s.%s.txt", string, o->name, mode);
   1392 
   1393 	log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode);
   1394 	ret = system(str);
   1395 	if (ret == -1)
   1396 		log_err("fio: exec of cmd <%s> failed\n", str);
   1397 
   1398 	free(str);
   1399 	return ret;
   1400 }
   1401 
   1402 /*
   1403  * Dry run to compute correct state of numberio for verification.
   1404  */
   1405 static uint64_t do_dry_run(struct thread_data *td)
   1406 {
   1407 	td_set_runstate(td, TD_RUNNING);
   1408 
   1409 	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
   1410 		(!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) {
   1411 		struct io_u *io_u;
   1412 		int ret;
   1413 
   1414 		if (td->terminate || td->done)
   1415 			break;
   1416 
   1417 		io_u = get_io_u(td);
   1418 		if (IS_ERR_OR_NULL(io_u))
   1419 			break;
   1420 
   1421 		io_u_set(td, io_u, IO_U_F_FLIGHT);
   1422 		io_u->error = 0;
   1423 		io_u->resid = 0;
   1424 		if (ddir_rw(acct_ddir(io_u)))
   1425 			td->io_issues[acct_ddir(io_u)]++;
   1426 		if (ddir_rw(io_u->ddir)) {
   1427 			io_u_mark_depth(td, 1);
   1428 			td->ts.total_io_u[io_u->ddir]++;
   1429 		}
   1430 
   1431 		if (td_write(td) && io_u->ddir == DDIR_WRITE &&
   1432 		    td->o.do_verify &&
   1433 		    td->o.verify != VERIFY_NONE &&
   1434 		    !td->o.experimental_verify)
   1435 			log_io_piece(td, io_u);
   1436 
   1437 		ret = io_u_sync_complete(td, io_u);
   1438 		(void) ret;
   1439 	}
   1440 
   1441 	return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
   1442 }
   1443 
   1444 struct fork_data {
   1445 	struct thread_data *td;
   1446 	struct sk_out *sk_out;
   1447 };
   1448 
   1449 /*
   1450  * Entry point for the thread based jobs. The process based jobs end up
   1451  * here as well, after a little setup.
   1452  */
   1453 static void *thread_main(void *data)
   1454 {
   1455 	struct fork_data *fd = data;
   1456 	unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, };
   1457 	struct thread_data *td = fd->td;
   1458 	struct thread_options *o = &td->o;
   1459 	struct sk_out *sk_out = fd->sk_out;
   1460 	uint64_t bytes_done[DDIR_RWDIR_CNT];
   1461 	int deadlock_loop_cnt;
   1462 	int clear_state;
   1463 	int ret;
   1464 
   1465 	sk_out_assign(sk_out);
   1466 	free(fd);
   1467 
   1468 	if (!o->use_thread) {
   1469 		setsid();
   1470 		td->pid = getpid();
   1471 	} else
   1472 		td->pid = gettid();
   1473 
   1474 	fio_local_clock_init(o->use_thread);
   1475 
   1476 	dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
   1477 
   1478 	if (is_backend)
   1479 		fio_server_send_start(td);
   1480 
   1481 	INIT_FLIST_HEAD(&td->io_log_list);
   1482 	INIT_FLIST_HEAD(&td->io_hist_list);
   1483 	INIT_FLIST_HEAD(&td->verify_list);
   1484 	INIT_FLIST_HEAD(&td->trim_list);
   1485 	INIT_FLIST_HEAD(&td->next_rand_list);
   1486 	td->io_hist_tree = RB_ROOT;
   1487 
   1488 	ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond);
   1489 	if (ret) {
   1490 		td_verror(td, ret, "mutex_cond_init_pshared");
   1491 		goto err;
   1492 	}
   1493 	ret = cond_init_pshared(&td->verify_cond);
   1494 	if (ret) {
   1495 		td_verror(td, ret, "mutex_cond_pshared");
   1496 		goto err;
   1497 	}
   1498 
   1499 	td_set_runstate(td, TD_INITIALIZED);
   1500 	dprint(FD_MUTEX, "up startup_mutex\n");
   1501 	fio_mutex_up(startup_mutex);
   1502 	dprint(FD_MUTEX, "wait on td->mutex\n");
   1503 	fio_mutex_down(td->mutex);
   1504 	dprint(FD_MUTEX, "done waiting on td->mutex\n");
   1505 
   1506 	/*
   1507 	 * A new gid requires privilege, so we need to do this before setting
   1508 	 * the uid.
   1509 	 */
   1510 	if (o->gid != -1U && setgid(o->gid)) {
   1511 		td_verror(td, errno, "setgid");
   1512 		goto err;
   1513 	}
   1514 	if (o->uid != -1U && setuid(o->uid)) {
   1515 		td_verror(td, errno, "setuid");
   1516 		goto err;
   1517 	}
   1518 
   1519 	/*
   1520 	 * Do this early, we don't want the compress threads to be limited
   1521 	 * to the same CPUs as the IO workers. So do this before we set
   1522 	 * any potential CPU affinity
   1523 	 */
   1524 	if (iolog_compress_init(td, sk_out))
   1525 		goto err;
   1526 
   1527 	/*
   1528 	 * If we have a gettimeofday() thread, make sure we exclude that
   1529 	 * thread from this job
   1530 	 */
   1531 	if (o->gtod_cpu)
   1532 		fio_cpu_clear(&o->cpumask, o->gtod_cpu);
   1533 
   1534 	/*
   1535 	 * Set affinity first, in case it has an impact on the memory
   1536 	 * allocations.
   1537 	 */
   1538 	if (fio_option_is_set(o, cpumask)) {
   1539 		if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) {
   1540 			ret = fio_cpus_split(&o->cpumask, td->thread_number - 1);
   1541 			if (!ret) {
   1542 				log_err("fio: no CPUs set\n");
   1543 				log_err("fio: Try increasing number of available CPUs\n");
   1544 				td_verror(td, EINVAL, "cpus_split");
   1545 				goto err;
   1546 			}
   1547 		}
   1548 		ret = fio_setaffinity(td->pid, o->cpumask);
   1549 		if (ret == -1) {
   1550 			td_verror(td, errno, "cpu_set_affinity");
   1551 			goto err;
   1552 		}
   1553 	}
   1554 
   1555 #ifdef CONFIG_LIBNUMA
   1556 	/* numa node setup */
   1557 	if (fio_option_is_set(o, numa_cpunodes) ||
   1558 	    fio_option_is_set(o, numa_memnodes)) {
   1559 		struct bitmask *mask;
   1560 
   1561 		if (numa_available() < 0) {
   1562 			td_verror(td, errno, "Does not support NUMA API\n");
   1563 			goto err;
   1564 		}
   1565 
   1566 		if (fio_option_is_set(o, numa_cpunodes)) {
   1567 			mask = numa_parse_nodestring(o->numa_cpunodes);
   1568 			ret = numa_run_on_node_mask(mask);
   1569 			numa_free_nodemask(mask);
   1570 			if (ret == -1) {
   1571 				td_verror(td, errno, \
   1572 					"numa_run_on_node_mask failed\n");
   1573 				goto err;
   1574 			}
   1575 		}
   1576 
   1577 		if (fio_option_is_set(o, numa_memnodes)) {
   1578 			mask = NULL;
   1579 			if (o->numa_memnodes)
   1580 				mask = numa_parse_nodestring(o->numa_memnodes);
   1581 
   1582 			switch (o->numa_mem_mode) {
   1583 			case MPOL_INTERLEAVE:
   1584 				numa_set_interleave_mask(mask);
   1585 				break;
   1586 			case MPOL_BIND:
   1587 				numa_set_membind(mask);
   1588 				break;
   1589 			case MPOL_LOCAL:
   1590 				numa_set_localalloc();
   1591 				break;
   1592 			case MPOL_PREFERRED:
   1593 				numa_set_preferred(o->numa_mem_prefer_node);
   1594 				break;
   1595 			case MPOL_DEFAULT:
   1596 			default:
   1597 				break;
   1598 			}
   1599 
   1600 			if (mask)
   1601 				numa_free_nodemask(mask);
   1602 
   1603 		}
   1604 	}
   1605 #endif
   1606 
   1607 	if (fio_pin_memory(td))
   1608 		goto err;
   1609 
   1610 	/*
   1611 	 * May alter parameters that init_io_u() will use, so we need to
   1612 	 * do this first.
   1613 	 */
   1614 	if (init_iolog(td))
   1615 		goto err;
   1616 
   1617 	if (init_io_u(td))
   1618 		goto err;
   1619 
   1620 	if (o->verify_async && verify_async_init(td))
   1621 		goto err;
   1622 
   1623 	if (fio_option_is_set(o, ioprio) ||
   1624 	    fio_option_is_set(o, ioprio_class)) {
   1625 		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
   1626 		if (ret == -1) {
   1627 			td_verror(td, errno, "ioprio_set");
   1628 			goto err;
   1629 		}
   1630 	}
   1631 
   1632 	if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt))
   1633 		goto err;
   1634 
   1635 	errno = 0;
   1636 	if (nice(o->nice) == -1 && errno != 0) {
   1637 		td_verror(td, errno, "nice");
   1638 		goto err;
   1639 	}
   1640 
   1641 	if (o->ioscheduler && switch_ioscheduler(td))
   1642 		goto err;
   1643 
   1644 	if (!o->create_serialize && setup_files(td))
   1645 		goto err;
   1646 
   1647 	if (td_io_init(td))
   1648 		goto err;
   1649 
   1650 	if (init_random_map(td))
   1651 		goto err;
   1652 
   1653 	if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun"))
   1654 		goto err;
   1655 
   1656 	if (o->pre_read) {
   1657 		if (pre_read_files(td) < 0)
   1658 			goto err;
   1659 	}
   1660 
   1661 	fio_verify_init(td);
   1662 
   1663 	if (rate_submit_init(td, sk_out))
   1664 		goto err;
   1665 
   1666 	set_epoch_time(td, o->log_unix_epoch);
   1667 	fio_getrusage(&td->ru_start);
   1668 	memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch));
   1669 	memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch));
   1670 	memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch));
   1671 
   1672 	if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
   1673 			o->ratemin[DDIR_TRIM]) {
   1674 	        memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
   1675 					sizeof(td->bw_sample_time));
   1676 	        memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
   1677 					sizeof(td->bw_sample_time));
   1678 	        memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
   1679 					sizeof(td->bw_sample_time));
   1680 	}
   1681 
   1682 	memset(bytes_done, 0, sizeof(bytes_done));
   1683 	clear_state = 0;
   1684 
   1685 	while (keep_running(td)) {
   1686 		uint64_t verify_bytes;
   1687 
   1688 		fio_gettime(&td->start, NULL);
   1689 		memcpy(&td->tv_cache, &td->start, sizeof(td->start));
   1690 
   1691 		if (clear_state) {
   1692 			clear_io_state(td, 0);
   1693 
   1694 			if (o->unlink_each_loop && unlink_all_files(td))
   1695 				break;
   1696 		}
   1697 
   1698 		prune_io_piece_log(td);
   1699 
   1700 		if (td->o.verify_only && td_write(td))
   1701 			verify_bytes = do_dry_run(td);
   1702 		else {
   1703 			do_io(td, bytes_done);
   1704 
   1705 			if (!ddir_rw_sum(bytes_done)) {
   1706 				fio_mark_td_terminate(td);
   1707 				verify_bytes = 0;
   1708 			} else {
   1709 				verify_bytes = bytes_done[DDIR_WRITE] +
   1710 						bytes_done[DDIR_TRIM];
   1711 			}
   1712 		}
   1713 
   1714 		/*
   1715 		 * If we took too long to shut down, the main thread could
   1716 		 * already consider us reaped/exited. If that happens, break
   1717 		 * out and clean up.
   1718 		 */
   1719 		if (td->runstate >= TD_EXITED)
   1720 			break;
   1721 
   1722 		clear_state = 1;
   1723 
   1724 		/*
   1725 		 * Make sure we've successfully updated the rusage stats
   1726 		 * before waiting on the stat mutex. Otherwise we could have
   1727 		 * the stat thread holding stat mutex and waiting for
   1728 		 * the rusage_sem, which would never get upped because
   1729 		 * this thread is waiting for the stat mutex.
   1730 		 */
   1731 		deadlock_loop_cnt = 0;
   1732 		do {
   1733 			check_update_rusage(td);
   1734 			if (!fio_mutex_down_trylock(stat_mutex))
   1735 				break;
   1736 			usleep(1000);
   1737 			if (deadlock_loop_cnt++ > 5000) {
   1738 				log_err("fio seems to be stuck grabbing stat_mutex, forcibly exiting\n");
   1739 				td->error = EDEADLK;
   1740 				goto err;
   1741 			}
   1742 		} while (1);
   1743 
   1744 		if (td_read(td) && td->io_bytes[DDIR_READ])
   1745 			update_runtime(td, elapsed_us, DDIR_READ);
   1746 		if (td_write(td) && td->io_bytes[DDIR_WRITE])
   1747 			update_runtime(td, elapsed_us, DDIR_WRITE);
   1748 		if (td_trim(td) && td->io_bytes[DDIR_TRIM])
   1749 			update_runtime(td, elapsed_us, DDIR_TRIM);
   1750 		fio_gettime(&td->start, NULL);
   1751 		fio_mutex_up(stat_mutex);
   1752 
   1753 		if (td->error || td->terminate)
   1754 			break;
   1755 
   1756 		if (!o->do_verify ||
   1757 		    o->verify == VERIFY_NONE ||
   1758 		    td_ioengine_flagged(td, FIO_UNIDIR))
   1759 			continue;
   1760 
   1761 		clear_io_state(td, 0);
   1762 
   1763 		fio_gettime(&td->start, NULL);
   1764 
   1765 		do_verify(td, verify_bytes);
   1766 
   1767 		/*
   1768 		 * See comment further up for why this is done here.
   1769 		 */
   1770 		check_update_rusage(td);
   1771 
   1772 		fio_mutex_down(stat_mutex);
   1773 		update_runtime(td, elapsed_us, DDIR_READ);
   1774 		fio_gettime(&td->start, NULL);
   1775 		fio_mutex_up(stat_mutex);
   1776 
   1777 		if (td->error || td->terminate)
   1778 			break;
   1779 	}
   1780 
   1781 	/*
   1782 	 * If td ended up with no I/O when it should have had,
   1783 	 * then something went wrong unless FIO_NOIO or FIO_DISKLESSIO.
   1784 	 * (Are we not missing other flags that can be ignored ?)
   1785 	 */
   1786 	if ((td->o.size || td->o.io_size) && !ddir_rw_sum(bytes_done) &&
   1787 	    !(td_ioengine_flagged(td, FIO_NOIO) ||
   1788 	      td_ioengine_flagged(td, FIO_DISKLESSIO)))
   1789 		log_err("%s: No I/O performed by %s, "
   1790 			 "perhaps try --debug=io option for details?\n",
   1791 			 td->o.name, td->io_ops->name);
   1792 
   1793 	td_set_runstate(td, TD_FINISHING);
   1794 
   1795 	update_rusage_stat(td);
   1796 	td->ts.total_run_time = mtime_since_now(&td->epoch);
   1797 	td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
   1798 	td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
   1799 	td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
   1800 
   1801 	if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) &&
   1802 	    (td->o.verify != VERIFY_NONE && td_write(td)))
   1803 		verify_save_state(td->thread_number);
   1804 
   1805 	fio_unpin_memory(td);
   1806 
   1807 	td_writeout_logs(td, true);
   1808 
   1809 	iolog_compress_exit(td);
   1810 	rate_submit_exit(td);
   1811 
   1812 	if (o->exec_postrun)
   1813 		exec_string(o, o->exec_postrun, (const char *)"postrun");
   1814 
   1815 	if (exitall_on_terminate || (o->exitall_error && td->error))
   1816 		fio_terminate_threads(td->groupid);
   1817 
   1818 err:
   1819 	if (td->error)
   1820 		log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
   1821 							td->verror);
   1822 
   1823 	if (o->verify_async)
   1824 		verify_async_exit(td);
   1825 
   1826 	close_and_free_files(td);
   1827 	cleanup_io_u(td);
   1828 	close_ioengine(td);
   1829 	cgroup_shutdown(td, &cgroup_mnt);
   1830 	verify_free_state(td);
   1831 
   1832 	if (td->zone_state_index) {
   1833 		int i;
   1834 
   1835 		for (i = 0; i < DDIR_RWDIR_CNT; i++)
   1836 			free(td->zone_state_index[i]);
   1837 		free(td->zone_state_index);
   1838 		td->zone_state_index = NULL;
   1839 	}
   1840 
   1841 	if (fio_option_is_set(o, cpumask)) {
   1842 		ret = fio_cpuset_exit(&o->cpumask);
   1843 		if (ret)
   1844 			td_verror(td, ret, "fio_cpuset_exit");
   1845 	}
   1846 
   1847 	/*
   1848 	 * do this very late, it will log file closing as well
   1849 	 */
   1850 	if (o->write_iolog_file)
   1851 		write_iolog_close(td);
   1852 
   1853 	td_set_runstate(td, TD_EXITED);
   1854 
   1855 	/*
   1856 	 * Do this last after setting our runstate to exited, so we
   1857 	 * know that the stat thread is signaled.
   1858 	 */
   1859 	check_update_rusage(td);
   1860 
   1861 	sk_out_drop();
   1862 	return (void *) (uintptr_t) td->error;
   1863 }
   1864 
   1865 /*
   1866  * Run over the job map and reap the threads that have exited, if any.
   1867  */
   1868 static void reap_threads(unsigned int *nr_running, uint64_t *t_rate,
   1869 			 uint64_t *m_rate)
   1870 {
   1871 	struct thread_data *td;
   1872 	unsigned int cputhreads, realthreads, pending;
   1873 	int i, status, ret;
   1874 
   1875 	/*
   1876 	 * reap exited threads (TD_EXITED -> TD_REAPED)
   1877 	 */
   1878 	realthreads = pending = cputhreads = 0;
   1879 	for_each_td(td, i) {
   1880 		int flags = 0;
   1881 
   1882 		/*
   1883 		 * ->io_ops is NULL for a thread that has closed its
   1884 		 * io engine
   1885 		 */
   1886 		if (td->io_ops && !strcmp(td->io_ops->name, "cpuio"))
   1887 			cputhreads++;
   1888 		else
   1889 			realthreads++;
   1890 
   1891 		if (!td->pid) {
   1892 			pending++;
   1893 			continue;
   1894 		}
   1895 		if (td->runstate == TD_REAPED)
   1896 			continue;
   1897 		if (td->o.use_thread) {
   1898 			if (td->runstate == TD_EXITED) {
   1899 				td_set_runstate(td, TD_REAPED);
   1900 				goto reaped;
   1901 			}
   1902 			continue;
   1903 		}
   1904 
   1905 		flags = WNOHANG;
   1906 		if (td->runstate == TD_EXITED)
   1907 			flags = 0;
   1908 
   1909 		/*
   1910 		 * check if someone quit or got killed in an unusual way
   1911 		 */
   1912 		ret = waitpid(td->pid, &status, flags);
   1913 		if (ret < 0) {
   1914 			if (errno == ECHILD) {
   1915 				log_err("fio: pid=%d disappeared %d\n",
   1916 						(int) td->pid, td->runstate);
   1917 				td->sig = ECHILD;
   1918 				td_set_runstate(td, TD_REAPED);
   1919 				goto reaped;
   1920 			}
   1921 			perror("waitpid");
   1922 		} else if (ret == td->pid) {
   1923 			if (WIFSIGNALED(status)) {
   1924 				int sig = WTERMSIG(status);
   1925 
   1926 				if (sig != SIGTERM && sig != SIGUSR2)
   1927 					log_err("fio: pid=%d, got signal=%d\n",
   1928 							(int) td->pid, sig);
   1929 				td->sig = sig;
   1930 				td_set_runstate(td, TD_REAPED);
   1931 				goto reaped;
   1932 			}
   1933 			if (WIFEXITED(status)) {
   1934 				if (WEXITSTATUS(status) && !td->error)
   1935 					td->error = WEXITSTATUS(status);
   1936 
   1937 				td_set_runstate(td, TD_REAPED);
   1938 				goto reaped;
   1939 			}
   1940 		}
   1941 
   1942 		/*
   1943 		 * If the job is stuck, do a forceful timeout of it and
   1944 		 * move on.
   1945 		 */
   1946 		if (td->terminate &&
   1947 		    td->runstate < TD_FSYNCING &&
   1948 		    time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
   1949 			log_err("fio: job '%s' (state=%d) hasn't exited in "
   1950 				"%lu seconds, it appears to be stuck. Doing "
   1951 				"forceful exit of this job.\n",
   1952 				td->o.name, td->runstate,
   1953 				(unsigned long) time_since_now(&td->terminate_time));
   1954 			td_set_runstate(td, TD_REAPED);
   1955 			goto reaped;
   1956 		}
   1957 
   1958 		/*
   1959 		 * thread is not dead, continue
   1960 		 */
   1961 		pending++;
   1962 		continue;
   1963 reaped:
   1964 		(*nr_running)--;
   1965 		(*m_rate) -= ddir_rw_sum(td->o.ratemin);
   1966 		(*t_rate) -= ddir_rw_sum(td->o.rate);
   1967 		if (!td->pid)
   1968 			pending--;
   1969 
   1970 		if (td->error)
   1971 			exit_value++;
   1972 
   1973 		done_secs += mtime_since_now(&td->epoch) / 1000;
   1974 		profile_td_exit(td);
   1975 	}
   1976 
   1977 	if (*nr_running == cputhreads && !pending && realthreads)
   1978 		fio_terminate_threads(TERMINATE_ALL);
   1979 }
   1980 
   1981 static bool __check_trigger_file(void)
   1982 {
   1983 	struct stat sb;
   1984 
   1985 	if (!trigger_file)
   1986 		return false;
   1987 
   1988 	if (stat(trigger_file, &sb))
   1989 		return false;
   1990 
   1991 	if (unlink(trigger_file) < 0)
   1992 		log_err("fio: failed to unlink %s: %s\n", trigger_file,
   1993 							strerror(errno));
   1994 
   1995 	return true;
   1996 }
   1997 
   1998 static bool trigger_timedout(void)
   1999 {
   2000 	if (trigger_timeout)
   2001 		return time_since_genesis() >= trigger_timeout;
   2002 
   2003 	return false;
   2004 }
   2005 
   2006 void exec_trigger(const char *cmd)
   2007 {
   2008 	int ret;
   2009 
   2010 	if (!cmd)
   2011 		return;
   2012 
   2013 	ret = system(cmd);
   2014 	if (ret == -1)
   2015 		log_err("fio: failed executing %s trigger\n", cmd);
   2016 }
   2017 
   2018 void check_trigger_file(void)
   2019 {
   2020 	if (__check_trigger_file() || trigger_timedout()) {
   2021 		if (nr_clients)
   2022 			fio_clients_send_trigger(trigger_remote_cmd);
   2023 		else {
   2024 			verify_save_state(IO_LIST_ALL);
   2025 			fio_terminate_threads(TERMINATE_ALL);
   2026 			exec_trigger(trigger_cmd);
   2027 		}
   2028 	}
   2029 }
   2030 
   2031 static int fio_verify_load_state(struct thread_data *td)
   2032 {
   2033 	int ret;
   2034 
   2035 	if (!td->o.verify_state)
   2036 		return 0;
   2037 
   2038 	if (is_backend) {
   2039 		void *data;
   2040 
   2041 		ret = fio_server_get_verify_state(td->o.name,
   2042 					td->thread_number - 1, &data);
   2043 		if (!ret)
   2044 			verify_assign_state(td, data);
   2045 	} else
   2046 		ret = verify_load_state(td, "local");
   2047 
   2048 	return ret;
   2049 }
   2050 
   2051 static void do_usleep(unsigned int usecs)
   2052 {
   2053 	check_for_running_stats();
   2054 	check_trigger_file();
   2055 	usleep(usecs);
   2056 }
   2057 
   2058 static bool check_mount_writes(struct thread_data *td)
   2059 {
   2060 	struct fio_file *f;
   2061 	unsigned int i;
   2062 
   2063 	if (!td_write(td) || td->o.allow_mounted_write)
   2064 		return false;
   2065 
   2066 	/*
   2067 	 * If FIO_HAVE_CHARDEV_SIZE is defined, it's likely that chrdevs
   2068 	 * are mkfs'd and mounted.
   2069 	 */
   2070 	for_each_file(td, f, i) {
   2071 #ifdef FIO_HAVE_CHARDEV_SIZE
   2072 		if (f->filetype != FIO_TYPE_BLOCK && f->filetype != FIO_TYPE_CHAR)
   2073 #else
   2074 		if (f->filetype != FIO_TYPE_BLOCK)
   2075 #endif
   2076 			continue;
   2077 		if (device_is_mounted(f->file_name))
   2078 			goto mounted;
   2079 	}
   2080 
   2081 	return false;
   2082 mounted:
   2083 	log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.\n", f->file_name);
   2084 	return true;
   2085 }
   2086 
   2087 static bool waitee_running(struct thread_data *me)
   2088 {
   2089 	const char *waitee = me->o.wait_for;
   2090 	const char *self = me->o.name;
   2091 	struct thread_data *td;
   2092 	int i;
   2093 
   2094 	if (!waitee)
   2095 		return false;
   2096 
   2097 	for_each_td(td, i) {
   2098 		if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee))
   2099 			continue;
   2100 
   2101 		if (td->runstate < TD_EXITED) {
   2102 			dprint(FD_PROCESS, "%s fenced by %s(%s)\n",
   2103 					self, td->o.name,
   2104 					runstate_to_name(td->runstate));
   2105 			return true;
   2106 		}
   2107 	}
   2108 
   2109 	dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee);
   2110 	return false;
   2111 }
   2112 
   2113 /*
   2114  * Main function for kicking off and reaping jobs, as needed.
   2115  */
   2116 static void run_threads(struct sk_out *sk_out)
   2117 {
   2118 	struct thread_data *td;
   2119 	unsigned int i, todo, nr_running, nr_started;
   2120 	uint64_t m_rate, t_rate;
   2121 	uint64_t spent;
   2122 
   2123 	if (fio_gtod_offload && fio_start_gtod_thread())
   2124 		return;
   2125 
   2126 	fio_idle_prof_init();
   2127 
   2128 	set_sig_handlers();
   2129 
   2130 	nr_thread = nr_process = 0;
   2131 	for_each_td(td, i) {
   2132 		if (check_mount_writes(td))
   2133 			return;
   2134 		if (td->o.use_thread)
   2135 			nr_thread++;
   2136 		else
   2137 			nr_process++;
   2138 	}
   2139 
   2140 	if (output_format & FIO_OUTPUT_NORMAL) {
   2141 		log_info("Starting ");
   2142 		if (nr_thread)
   2143 			log_info("%d thread%s", nr_thread,
   2144 						nr_thread > 1 ? "s" : "");
   2145 		if (nr_process) {
   2146 			if (nr_thread)
   2147 				log_info(" and ");
   2148 			log_info("%d process%s", nr_process,
   2149 						nr_process > 1 ? "es" : "");
   2150 		}
   2151 		log_info("\n");
   2152 		log_info_flush();
   2153 	}
   2154 
   2155 	todo = thread_number;
   2156 	nr_running = 0;
   2157 	nr_started = 0;
   2158 	m_rate = t_rate = 0;
   2159 
   2160 	for_each_td(td, i) {
   2161 		print_status_init(td->thread_number - 1);
   2162 
   2163 		if (!td->o.create_serialize)
   2164 			continue;
   2165 
   2166 		if (fio_verify_load_state(td))
   2167 			goto reap;
   2168 
   2169 		/*
   2170 		 * do file setup here so it happens sequentially,
   2171 		 * we don't want X number of threads getting their
   2172 		 * client data interspersed on disk
   2173 		 */
   2174 		if (setup_files(td)) {
   2175 reap:
   2176 			exit_value++;
   2177 			if (td->error)
   2178 				log_err("fio: pid=%d, err=%d/%s\n",
   2179 					(int) td->pid, td->error, td->verror);
   2180 			td_set_runstate(td, TD_REAPED);
   2181 			todo--;
   2182 		} else {
   2183 			struct fio_file *f;
   2184 			unsigned int j;
   2185 
   2186 			/*
   2187 			 * for sharing to work, each job must always open
   2188 			 * its own files. so close them, if we opened them
   2189 			 * for creation
   2190 			 */
   2191 			for_each_file(td, f, j) {
   2192 				if (fio_file_open(f))
   2193 					td_io_close_file(td, f);
   2194 			}
   2195 		}
   2196 	}
   2197 
   2198 	/* start idle threads before io threads start to run */
   2199 	fio_idle_prof_start();
   2200 
   2201 	set_genesis_time();
   2202 
   2203 	while (todo) {
   2204 		struct thread_data *map[REAL_MAX_JOBS];
   2205 		struct timeval this_start;
   2206 		int this_jobs = 0, left;
   2207 		struct fork_data *fd;
   2208 
   2209 		/*
   2210 		 * create threads (TD_NOT_CREATED -> TD_CREATED)
   2211 		 */
   2212 		for_each_td(td, i) {
   2213 			if (td->runstate != TD_NOT_CREATED)
   2214 				continue;
   2215 
   2216 			/*
   2217 			 * never got a chance to start, killed by other
   2218 			 * thread for some reason
   2219 			 */
   2220 			if (td->terminate) {
   2221 				todo--;
   2222 				continue;
   2223 			}
   2224 
   2225 			if (td->o.start_delay) {
   2226 				spent = utime_since_genesis();
   2227 
   2228 				if (td->o.start_delay > spent)
   2229 					continue;
   2230 			}
   2231 
   2232 			if (td->o.stonewall && (nr_started || nr_running)) {
   2233 				dprint(FD_PROCESS, "%s: stonewall wait\n",
   2234 							td->o.name);
   2235 				break;
   2236 			}
   2237 
   2238 			if (waitee_running(td)) {
   2239 				dprint(FD_PROCESS, "%s: waiting for %s\n",
   2240 						td->o.name, td->o.wait_for);
   2241 				continue;
   2242 			}
   2243 
   2244 			init_disk_util(td);
   2245 
   2246 			td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
   2247 			td->update_rusage = 0;
   2248 
   2249 			/*
   2250 			 * Set state to created. Thread will transition
   2251 			 * to TD_INITIALIZED when it's done setting up.
   2252 			 */
   2253 			td_set_runstate(td, TD_CREATED);
   2254 			map[this_jobs++] = td;
   2255 			nr_started++;
   2256 
   2257 			fd = calloc(1, sizeof(*fd));
   2258 			fd->td = td;
   2259 			fd->sk_out = sk_out;
   2260 
   2261 			if (td->o.use_thread) {
   2262 				int ret;
   2263 
   2264 				dprint(FD_PROCESS, "will pthread_create\n");
   2265 				ret = pthread_create(&td->thread, NULL,
   2266 							thread_main, fd);
   2267 				if (ret) {
   2268 					log_err("pthread_create: %s\n",
   2269 							strerror(ret));
   2270 					free(fd);
   2271 					nr_started--;
   2272 					break;
   2273 				}
   2274 				ret = pthread_detach(td->thread);
   2275 				if (ret)
   2276 					log_err("pthread_detach: %s",
   2277 							strerror(ret));
   2278 			} else {
   2279 				pid_t pid;
   2280 				dprint(FD_PROCESS, "will fork\n");
   2281 				pid = fork();
   2282 				if (!pid) {
   2283 					int ret;
   2284 
   2285 					ret = (int)(uintptr_t)thread_main(fd);
   2286 					_exit(ret);
   2287 				} else if (i == fio_debug_jobno)
   2288 					*fio_debug_jobp = pid;
   2289 			}
   2290 			dprint(FD_MUTEX, "wait on startup_mutex\n");
   2291 			if (fio_mutex_down_timeout(startup_mutex, 10000)) {
   2292 				log_err("fio: job startup hung? exiting.\n");
   2293 				fio_terminate_threads(TERMINATE_ALL);
   2294 				fio_abort = 1;
   2295 				nr_started--;
   2296 				break;
   2297 			}
   2298 			dprint(FD_MUTEX, "done waiting on startup_mutex\n");
   2299 		}
   2300 
   2301 		/*
   2302 		 * Wait for the started threads to transition to
   2303 		 * TD_INITIALIZED.
   2304 		 */
   2305 		fio_gettime(&this_start, NULL);
   2306 		left = this_jobs;
   2307 		while (left && !fio_abort) {
   2308 			if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
   2309 				break;
   2310 
   2311 			do_usleep(100000);
   2312 
   2313 			for (i = 0; i < this_jobs; i++) {
   2314 				td = map[i];
   2315 				if (!td)
   2316 					continue;
   2317 				if (td->runstate == TD_INITIALIZED) {
   2318 					map[i] = NULL;
   2319 					left--;
   2320 				} else if (td->runstate >= TD_EXITED) {
   2321 					map[i] = NULL;
   2322 					left--;
   2323 					todo--;
   2324 					nr_running++; /* work-around... */
   2325 				}
   2326 			}
   2327 		}
   2328 
   2329 		if (left) {
   2330 			log_err("fio: %d job%s failed to start\n", left,
   2331 					left > 1 ? "s" : "");
   2332 			for (i = 0; i < this_jobs; i++) {
   2333 				td = map[i];
   2334 				if (!td)
   2335 					continue;
   2336 				kill(td->pid, SIGTERM);
   2337 			}
   2338 			break;
   2339 		}
   2340 
   2341 		/*
   2342 		 * start created threads (TD_INITIALIZED -> TD_RUNNING).
   2343 		 */
   2344 		for_each_td(td, i) {
   2345 			if (td->runstate != TD_INITIALIZED)
   2346 				continue;
   2347 
   2348 			if (in_ramp_time(td))
   2349 				td_set_runstate(td, TD_RAMP);
   2350 			else
   2351 				td_set_runstate(td, TD_RUNNING);
   2352 			nr_running++;
   2353 			nr_started--;
   2354 			m_rate += ddir_rw_sum(td->o.ratemin);
   2355 			t_rate += ddir_rw_sum(td->o.rate);
   2356 			todo--;
   2357 			fio_mutex_up(td->mutex);
   2358 		}
   2359 
   2360 		reap_threads(&nr_running, &t_rate, &m_rate);
   2361 
   2362 		if (todo)
   2363 			do_usleep(100000);
   2364 	}
   2365 
   2366 	while (nr_running) {
   2367 		reap_threads(&nr_running, &t_rate, &m_rate);
   2368 		do_usleep(10000);
   2369 	}
   2370 
   2371 	fio_idle_prof_stop();
   2372 
   2373 	update_io_ticks();
   2374 }
   2375 
   2376 static void free_disk_util(void)
   2377 {
   2378 	disk_util_prune_entries();
   2379 	helper_thread_destroy();
   2380 }
   2381 
   2382 int fio_backend(struct sk_out *sk_out)
   2383 {
   2384 	struct thread_data *td;
   2385 	int i;
   2386 
   2387 	if (exec_profile) {
   2388 		if (load_profile(exec_profile))
   2389 			return 1;
   2390 		free(exec_profile);
   2391 		exec_profile = NULL;
   2392 	}
   2393 	if (!thread_number)
   2394 		return 0;
   2395 
   2396 	if (write_bw_log) {
   2397 		struct log_params p = {
   2398 			.log_type = IO_LOG_TYPE_BW,
   2399 		};
   2400 
   2401 		setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log");
   2402 		setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log");
   2403 		setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
   2404 	}
   2405 
   2406 	startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
   2407 	if (startup_mutex == NULL)
   2408 		return 1;
   2409 
   2410 	set_genesis_time();
   2411 	stat_init();
   2412 	helper_thread_create(startup_mutex, sk_out);
   2413 
   2414 	cgroup_list = smalloc(sizeof(*cgroup_list));
   2415 	INIT_FLIST_HEAD(cgroup_list);
   2416 
   2417 	run_threads(sk_out);
   2418 
   2419 	helper_thread_exit();
   2420 
   2421 	if (!fio_abort) {
   2422 		__show_run_stats();
   2423 		if (write_bw_log) {
   2424 			for (i = 0; i < DDIR_RWDIR_CNT; i++) {
   2425 				struct io_log *log = agg_io_log[i];
   2426 
   2427 				flush_log(log, false);
   2428 				free_log(log);
   2429 			}
   2430 		}
   2431 	}
   2432 
   2433 	for_each_td(td, i) {
   2434 		if (td->ss.dur) {
   2435 			if (td->ss.iops_data != NULL) {
   2436 				free(td->ss.iops_data);
   2437 				free(td->ss.bw_data);
   2438 			}
   2439 		}
   2440 		fio_options_free(td);
   2441 		if (td->rusage_sem) {
   2442 			fio_mutex_remove(td->rusage_sem);
   2443 			td->rusage_sem = NULL;
   2444 		}
   2445 		fio_mutex_remove(td->mutex);
   2446 		td->mutex = NULL;
   2447 	}
   2448 
   2449 	free_disk_util();
   2450 	cgroup_kill(cgroup_list);
   2451 	sfree(cgroup_list);
   2452 	sfree(cgroup_mnt);
   2453 
   2454 	fio_mutex_remove(startup_mutex);
   2455 	stat_exit();
   2456 	return exit_value;
   2457 }
   2458